Quote:
Originally Posted by DNSB
Anyone seen the response in one tech publication to the report that OpenAI's own tests show that o3 and o4-mini have more issues with hallucinations than their earlier versions? I was not able to find it but the comment I saw elsewhere condensed the original into saying that LLMs exemplify the trust but verify philosophy. I suspect the link to the original text was been hallucinated by an AI.
|
And "hallucination" is marketing speak for "not even plausible nonsense". It's inherent. There is no difference between so called "hallucination" and "plausible nonsense" if it's a subject where you are very expert, because the "plausible nonsense" is only plausible if you don't know the stuff. Hence a failure at law, summaries of company meeting minutes, review of CVs or anything where the result affects revenue or people's lives.
I'm told there is no truth in the rumour of Copilot for Xbox to play your games while you do the chores.